RAG Is Broken — Build a Knowledge System

Infographic contrasting traditional RAG (Retrieval-Augmented Generation) with a structured Knowledge System / LLM Wiki, featuring Andrej Karpathy.

Core Argument

RAG = messy pile of PDFs, docs, spreadsheets with tangled retrieval. Fragmented, no context, shallow.

Knowledge System = structured network of Concepts, Entities, Overview, Insights, and Connections — a knowledge graph that compounds, stays consistent, and gets smarter over time.

Key Value Props

• Compound Knowledge — every piece connects and enriches others
• Stay Consistent — structured context over ad-hoc retrieval
• Get Smarter Over Time — the system improves with use

Relevance

Directly aligns with:

• ObsidianBrain — the iOS second brain app concept is exactly this: a structured knowledge system, not a RAG pipeline over raw files
• HermesVault / kb-compile pipeline — vault compilation into structured wiki notes mirrors the "knowledge system" approach
• Karpathy's LLM Wiki concept — which the user already follows via the karpathy-llm-wiki-obsidian skill